Deep Learning Tools for image classification in Cryo-electron microscopy

Abstract

Cryo-electron microscopy is an imaging technique used for 3D reconstruction of biomolecules, enabling researchers to study their structures. However, due to low signal-to-noise ratios in captured images, 2D classification is a critical preprocessing step. This thesis explores the application of a deep learning approach, specifically a similarity network, to address this challenge. A Siamese model, trained with a Triplet Loss function, is used to differentiate between similar and dissimilar images. The model was trained on a dataset with known ground truth and tested on two types of unseen data: a similar dataset with ground truth and a different dataset without the ground truth. This study demonstrates the potential of deep learning to complement traditional 2D classification methods in cryo-EM.